In computers and other complex electrical systems it can be useful for diagnostic purposes to maintain a record of various parameters such as temperature or supply voltage. This data can subsequently be used for investigating faults, sometimes on a statistical basis. For example, it might be seen that components that may be subjected to a lower temperature or operated at a somewhat higher voltage are more prone to certain types of fault. Such analysis can then be used to improve reliability by redesigning the components concerned and/or by modifying the recommended operating parameters.
A typical mechanism for collecting such information is to locate monitors at various positions within the system in order to measure the parameter(s) of interest. A central management system, such as a service processor, is then used to poll the monitors in order to collect and store the monitor readings. This polling is typically performed on a regular basis, say every few minutes, or perhaps every time a certain type of operation is invoked.
However, polling in this manner is liable to miss short-term excursions in a parameter being monitored. This is unfortunate, in that such excursions may be of particular interest from a diagnostic perspective. In addition, repeatedly polling a parameter that is (in fact) relatively constant may be regarded as an undesirable overhead on the general operation of the system.